HFSS v15 HPC problem using command line
时间：03-30
整理：3721RD
点击：
Hi,
I am trying to run simulations on HFSS which has been installed on a linux based HPC platform. There are 1000's of cores available for me to use. Each computing node has 16 cores with maximum 32GB memory per node. I am trying to simulate antennas around 2THz frequency range. I have requested for exclusive access to 2 complete nodes for my simulation, i.e., 32cores and 64GB memory in total. I have to submit my HFSS job via command line and the code is as follows:
hfss -distributed -monitor -machinelist num=32 -ng -batchoptions \
"HFSS/HPCLicenseType=Pool HFSS/NumCoresPerDistributedTask=32 HFSS/DesiredRAMLimitInGB=64 tempdirectory=S{TMPDIR}" \
-batchsolve <filename>.hfss
This does generate a log file which can be followed to check the progress of the simulation.
It solves few adaptive passes before converging, but only uses one core to do it.
Multiple cores are only used for frequency sweeps, but not always. If I have only few frequency steps, it just uses one core and does not distribute task to other available cores.
My simulation crashes as soon as one of the nodes exceeds 32GB, at the same point, even if I allocate 3 nodes, i.e., 32X3=96GB.
Am I not feeding the correct parameters for distributed task? I am sure HFSS does have MPI functionality. Do I have to specify it somewhere in the command line?
Please help me out. Thank you very much! :)
I am trying to run simulations on HFSS which has been installed on a linux based HPC platform. There are 1000's of cores available for me to use. Each computing node has 16 cores with maximum 32GB memory per node. I am trying to simulate antennas around 2THz frequency range. I have requested for exclusive access to 2 complete nodes for my simulation, i.e., 32cores and 64GB memory in total. I have to submit my HFSS job via command line and the code is as follows:
hfss -distributed -monitor -machinelist num=32 -ng -batchoptions \
"HFSS/HPCLicenseType=Pool HFSS/NumCoresPerDistributedTask=32 HFSS/DesiredRAMLimitInGB=64 tempdirectory=S{TMPDIR}" \
-batchsolve <filename>.hfss
This does generate a log file which can be followed to check the progress of the simulation.
It solves few adaptive passes before converging, but only uses one core to do it.
Multiple cores are only used for frequency sweeps, but not always. If I have only few frequency steps, it just uses one core and does not distribute task to other available cores.
My simulation crashes as soon as one of the nodes exceeds 32GB, at the same point, even if I allocate 3 nodes, i.e., 32X3=96GB.
Am I not feeding the correct parameters for distributed task? I am sure HFSS does have MPI functionality. Do I have to specify it somewhere in the command line?
Please help me out. Thank you very much! :)